TY - THES

T1 - Supervised learning algorithms to predict porosity and porous zones of the Mount Messenger Formation (Taranaki Basin, NZ) based on petrophysical analysis of geophysical borehole data

AU - Kink, Daniela

N1 - no embargo

PY - 2023

Y1 - 2023

N2 - The reduction of carbon emission to the atmosphere is one of the most important tasks in the world. One way to reduce carbon emission to achieve global goals on climate change is to increase the number of geothermal reservoirs and to increase the amount of reservoirs, where CO2 can be stored. It is time-consuming to evaluate the potential of these reservoirs to get a precise and fast estimation of porosity and porous zones as part of the reservoir characterization. One approach to accelerate this process is the use of machine learning algorithms. In this thesis a workflow is created to test such algorithms on a pilot area in the Mount Messenger Formation of the Taranaki Basin, New Zealand. In this thesis two kinds of machine learning models are used: a linear regression model to estimate the porosity and a classification model using an artificial neural network (ANN) to predict porous zones. The input features of the machine learning models are based on the petrophysical analysis (multimineral analysis) and on geophysical borehole data of five wells. The output feature for the regression model is the petrophysically estimated porosity and the output feature of the classification model is the interpretation of potential reservoir zones, classified by no potential or potential of the same five wells. The regression model and the classification model are evaluated by the coefficient of determination (R2) and the accuracy of the validation dataset, the results indicate that the combination of geophysical borehole logs and a multimineral analysis provide an accurate model by 99.81% and 98,44%. Both models were applied to new data of two other boreholes and are showing similar high accuracy. Nevertheless, the outcome of a good model is amongst other things dependent on the well log data provided, the accuracy of the multimineral analysis and all pre-processing steps ahead of any model building process. It is indispensable to choose the features and the model accordingly to the data available and the outcome of interest.

AB - The reduction of carbon emission to the atmosphere is one of the most important tasks in the world. One way to reduce carbon emission to achieve global goals on climate change is to increase the number of geothermal reservoirs and to increase the amount of reservoirs, where CO2 can be stored. It is time-consuming to evaluate the potential of these reservoirs to get a precise and fast estimation of porosity and porous zones as part of the reservoir characterization. One approach to accelerate this process is the use of machine learning algorithms. In this thesis a workflow is created to test such algorithms on a pilot area in the Mount Messenger Formation of the Taranaki Basin, New Zealand. In this thesis two kinds of machine learning models are used: a linear regression model to estimate the porosity and a classification model using an artificial neural network (ANN) to predict porous zones. The input features of the machine learning models are based on the petrophysical analysis (multimineral analysis) and on geophysical borehole data of five wells. The output feature for the regression model is the petrophysically estimated porosity and the output feature of the classification model is the interpretation of potential reservoir zones, classified by no potential or potential of the same five wells. The regression model and the classification model are evaluated by the coefficient of determination (R2) and the accuracy of the validation dataset, the results indicate that the combination of geophysical borehole logs and a multimineral analysis provide an accurate model by 99.81% and 98,44%. Both models were applied to new data of two other boreholes and are showing similar high accuracy. Nevertheless, the outcome of a good model is amongst other things dependent on the well log data provided, the accuracy of the multimineral analysis and all pre-processing steps ahead of any model building process. It is indispensable to choose the features and the model accordingly to the data available and the outcome of interest.

KW - supervised learning

KW - multiple linear regression

KW - deep learning

KW - Mount Messenger Formation

KW - Supervised Learning

KW - Lineare Regression

KW - Deep Learning

KW - Mount Messenger Formation

KW - Petrophysik

KW - Poröse Zonen

M3 - Master's Thesis

ER -